Leaning into STEM: Predicting STEM Major Choice Using ... - ACT

1

ACT Research & Policy | Technical Brief | October 2018

Leaning into STEM: Predicting STEM Major Choice Using Measures of Academic Tilt and Measured Interest Tilt Paul Westrick, PhD, Justine Radunzel, PhD, & Dina Bassiri, PhD This study examined the value of using measures of academic “tilt” and vocational interest “tilt” to predict whether students will declare a STEM major during their first term in college. Academic tilt looks at students’ relative academic strengths by appropriately comparing their math and science achievement levels to their English, reading, and social studies achievement levels. Vocational interest tilt measures are based on students’ People/Things and Data/Ideas work-task dimension scores that underlie the ACT World-ofWork map. Results suggested that having a relative strength in math and science achievement and having a tilt toward things and ideas on the People/Things and Data/Ideas dimensions are positively related to STEM major choice, after statistically controlling for mathematics and science academic achievement levels, high school coursework taken and grades earned, major intentions, certainty of major intentions, and gender. As new initiatives and programs are

(Le, Robbins, & Westrick, 2014; Mattern &

implemented to promote STEM (Science,

Radunzel, 2016; Mattern, Radunzel, & Westrick,

Technology, Engineering, and Mathematics)

2015; Radunzel, Mattern, & Westrick, 2016).

interest and participation among US students, it is important to gain a better understanding

It is also widely recognized that female

of what student characteristics are useful for

students are underrepresented in math-

identifying those who are likely to declare a

intensive STEM fields, both in college and in the

STEM major during their first term in college.

workforce (National Science Foundation, 2015).

Prior research has found that having higher

Researchers have examined this issue from

mathematics and science standardized test

multiple perspectives and provided alternative

scores, taking higher-level mathematics and

explanations for the difference in male/female

science coursework in high school, expressing

participation rates. Some researchers have

interest in a STEM-related field, and having

studied the gap by examining vocational

measured vocational interests in STEM are

interests. Using Holland’s (1959; 1997)

positively related to STEM major choice,

framework of vocational interests and

persistence, and degree completion in college

Prediger’s (1982) People/Things (PT) and

ACT.org/research

ACT Research & Policy | Technical Brief | July 2018

2

Data/Ideas (DI) work-task dimensions, there is

generally showed greater strength in English

strong evidence that males prefer occupations

and reading on average across all the majors

that are associated with working with things

examined in the study, but the magnitude of

and females’ interests are more aligned with

the tilt toward higher verbal skills was found to

occupations that emphasize working with

be higher among verbal-intensive majors than

people (Su, Rounds, & Armstrong, 2009). These

among math-intensive majors with almost no

studies also suggest that math-intensive STEM

tilt found in Engineering. These findings

fields such as engineering tend to attract

highlight the need for further research on this

students with a stronger interest in working

topic to better understand whether students’

with things.

relative academic strengths help to explain why females tend to be underrepresented in math-

Other research suggests that students’

intensive STEM fields.

perceptions of their relative strength, whether they are better (or worse) at math-intensive

Building on these previous studies, the study

studies than they are at verbal-intensive

had three objectives. The objectives were to 1)

studies, influences students’ decisions on

develop measures of relative academic

whether or not to enroll, persist, earn a degree,

strength or “tilt” as related to differences in

and eventually work in a STEM field (Coyle,

STEM and non-STEM achievement levels; 2)

Purcell, Snyder, & Richmond, 2014; Davison,

examine whether academic tilt and established

Jew, & Davenport, 2014; Riegle-Crumb, King,

vocational interest tilt measures on Prediger’s

Grodsky, & Muller, 2012). The literature also

People/Things and Data/Ideas dimensions are

suggests that students with high mathematics

predictive of STEM major choice, after

ability also tend to have high verbal abilities,

statistically controlling for mathematics and

and this is especially true for female students,

science achievement, high school coursework

giving them a wide variety of academic majors

taken and grades earned, major intentions,

from which to choose (Lubinski & Benbow,

certainty of major intentions, and gender; and

2007). In contrast, male students with high

3) determine if the effects of the academic

mathematics ability are said to have a larger

achievement and tilt measures on STEM major

gap in their mathematics and verbal abilities

choice differ by gender.

and are thus more likely to see themselves as constrained to math-intensive STEM fields (e.g., Lubinski & Benbow, 2007). An exploratory analysis from a recent study by Westrick (2018) provides some support for the notion that students tend to gravitate toward academic majors that match their relative academic strengths. The study examined the precollege academic and measured interest profiles of fourth-year undergraduate students and found that males generally showed greater strength in math and science on average among those in STEM majors and greater strength in English and reading among those in non-STEM majors. In comparison, females

Data The data included nearly 80,000 students who took the ACT® test, enrolled as a first-time entering student in fall 2005 through 2009, and declared a major during their first fall term of enrollment. More than 40 four-year postsecondary institutions were represented in the sample. The institutions were diverse with regard to institutional control, selectivity, and size. Seventy percent were public institutions, 23% had highly selective admission policies, and more than one-half (56%) had a student body of fewer than 5,000 students. Institutions

R1700


provided the six-digit Classification of Instruction Program (CIP) major codes; these

3

Defining Academic Tilt

codes were used to identify STEM and non-

There are multiple ways that one might

STEM majors. Students’ demographic

compare test scores to define a relative

characteristics, high school coursework and

academic strength or “tilt” measure on

grades, major plans and interests, and ACT

differences in students’ STEM and non-STEM

Interest Inventory results were obtained from

achievement levels. Two methods include

the ACT registration database; students

examining (1) the difference in scale scores and

provided this information at the time they

(2) the difference in score percentiles. Each

registered to take the ACT. Official ACT test

approach has its weakness. For example, a

scores were also obtained. If students took the

student may have a higher scale score on Test

ACT more than once, only scores from the most

A than on Test B, but the student may have a

recent ACT administration were used.

higher percentile rank on Test B than on Test A. The problem is that score distributions can

Methods

differ across subject tests that use the same

Defining STEM Major Choice

student’s relative academic strength or “tilt”

The study outcome was math-intensive STEM

Standardizing the scale scores for each

major choice during the first fall term of college

measure eliminates this issue (Coyle et al., 2014;

enrollment and was coded as a binary outcome

Shea, Lubinski, & Benbow, 2001). Therefore, in

(STEM major (1) vs. non-STEM major (0)).

this study, relative academic strengths were

Researchers have distinguished between

determined by subtracting the z-scores of non-

math-intensive STEM fields, non-math-

STEM achievement from the z-scores of STEM

score scale. Based on these methods, the would depend on the approach used.

intensive STEM fields, and non-STEM fields (e.g.,

achievement. The means and standard

Ceci, Williams, & Barnett, 2009; Pennock-

deviations from a national reference group of

Roman, 1994). The math-intensive STEM

ACT-tested students were used to convert the

programs included in this study were Engineering and Technology, Computer Science and Mathematics, and Natural Science

scale scores to z-scores. ACT examinees from fall 2003 through spring 2009 served as the reference group for this study. This approach

majors.1 Inclusion of these majors as being

accounted for differences in the score

math-intensive is supported by the results from

distributions across subject areas.

a catalog and course transcript review of the typical first-year mathematics course taken in college by STEM majors (overall and by cluster) and non-STEM majors (see Table 3 from Mattern et al. (2015) for more details). For the purposes of this study, Medical & Health majors, which are sometimes considered STEM majors (ACT, 2016), were not included. All other majors were classified as non-STEM.

Two different relative academic strength variables that measured a tilt toward STEMrelated subject areas were examined in this study. The first variable (labeled as ACT tilt) was based on ACT test scores. It was calculated as the difference in the standardized mean ACT mathematics and science score (or the ACT STEM score) minus the standardized mean ACT English and reading score. The second variable (referred to as high school grade point average (GPA) tilt) was based on students’ grades earned in their high school courses. It was R1700


4

calculated as the difference in the standardized

major intentions (categorized as very sure, fairly

high school STEM GPA (that is, the mean high

sure, not sure); taking Calculus in high school

school mathematics and science GPA) minus

(yes=1, no=0); taking Physics in high school

the standardized mean high school English

(yes=1, no=0); and gender (male=1, female=0).

and social studies GPA.

Major sureness was relevant for students with an intended major only; it was not evaluated for

Defining Vocational Interest Tilt Students’ ACT Interest Inventory scores were converted to Prediger’s (1982) People/Things and Data/Ideas work-task dimension scores, which serve as ideal measures of vocational interest tilt. Students’ ACT Interest Inventory scores are reported visually on the ACT Worldof-Work Map (see Figure 1), which links individuals measured interests to career clusters. Underlying the map are the People/Things and Data/Ideas dimensions. Math-intensive STEM fields such as Engineering & Technologies, Natural Science & Technologies, and Computer & Information Specialties are located in the right of the map, oriented more toward Things than toward People on the People/Things dimension. Most STEM fields are located in the bottom half of the map, higher on Ideas than on Data on the Data/Ideas dimension, though Computer & Information Specialties are located near the middle of the map. Students’ People/Things and Data/Ideas work-task dimension scores were converted to z-scores using the same national reference group of ACT-tested students that was used when standardizing students’ ACT scores and subject area high school GPAs. The standardized z-scores served as the People/Things and Data/Ideas tilt measures in this study.

Other Predictors Other predictors included intended academic major (categorized as STEM, medical and health, non-STEM, and undecided); certainty of

undecided students. Consequently, when the major sureness variable was included as a predictor in the models, it was included in relation to intended major.2

Statistical Procedures Due to the nested structure of the data, hierarchical logistic regression models with random slopes and random intercepts were used to estimate students’ likelihood of declaring a math-intensive STEM major. Three models were compared. The first included measures of academic achievement, intended academic major, sureness of intended major, and gender. The second model added the four tilt measures of ACT-tilt, HSGPA-tilt, People/Things tilt, and Data/Ideas tilt to the first model, where the first two variables were the academic tilt measures and the last two were the vocational interest tilt measures. The third model added interactions between gender and six other predictors to the second model: ACT STEM score, STEM-HSGPA, ACT-tilt, HSGPA-tilt, People/Things tilt, and Data/Ideas tilt. To assess model fit, we calculated the typical accuracy rate (AR) and logistic R across institutions. The AR estimates the proportion of students correctly identified as entering either a math-intensive STEM major or a non-STEM major. The logistic R – defined as the standard deviation of the estimated logit function (Allen & Le, 2008) – measures the overall predictive strength of the model. The higher the logistic R is, the stronger the relationship is between the predictors and the criterion. This measure is derived in a manner analogous to that for the multiple R in multiple linear regression, but it is

R1700


5

appropriate for logistic regression models.

dimension in the sample and nationally were

Given that it is the standard deviation of the

smaller, with means close to zero for both

estimated logit function, it is not bounded

males and females.

between 0 and 1 as is the multiple R.

Results Descriptive Statistics Tables 1 and 2 contain descriptive statistics on student characteristics for the national population of ACT-tested students and the sample, respectively. The study sample tended to be better prepared academically than the national population, on average. For example, the means for ACT scores – ACT STEM and ACT English and reading – and high school GPAs – STEM HSGPA and English and Social Studies HSGPA – were higher in the sample than in the population. The directionality of the tilt measures by gender tended to be similar between the sample and the population. For instance, the mean ACT-tilt for male students was 0.18 nationally and 0.22 in the sample. For female students, the national mean was -0.14 compared to -0.33 in the sample. This indicated that male students had performed, on average, relatively stronger on ACT math and science tests than they had on the English and reading tests, and female students had performed, on average, relatively stronger on the ACT English and reading tests than they had on the math and science tests. A similar pattern held for HSGPA tilt where the standardized value was 0.07 for male students and -0.07 for female students, nationally, but the mean HSGPA-tilt in the sample was 0.05 and 0.03 for male and female students, respectively. For measured interests, the male average on the People/Things dimension was 0.29 nationally and 0.21 in the sample (tilted toward Things) whereas the female average was -0.23 nationally and -0.33 in the sample (tilted toward People). Gender differences on the Data/Ideas

Predicting STEM Major Choice Overall, 34% of the students in the sample declared a math-intensive STEM major during their first term in college; however, this rate varied across institutions (median = 30%, 10th percentile = 17%, 90th percentile = 39%). Additionally, the likelihood of declaring a mathintensive STEM degree was found to be significantly related to all of the student characteristics included in the first model (see Table 3). Specifically, students who entered college better prepared academically in mathematics and science had a greater likelihood of declaring a math-intensive STEM major during their first semester. This is evidenced by all of the academic measures – ACT STEM score, STEM HSGPA, taking a Calculus course, and taking a Physics course in high school – being positively related to the likelihood of declaring a math-intensive STEM major. Additionally, students who had expressed interest in a math-intensive STEM major or a Medical & Health STEM major, or who were undecided about their major intentions were more likely to declare a mathintensive STEM major during their first semester than those who had planned to major in a non-STEM field. Moreover, the likelihood of declaring a STEM major increased as a student’s certainty about their STEM major intentions increased. The effect for gender was positive, indicating that male students were more likely to declare a math-intensive STEM major than were female students (adjusted odds ratio = 1.39), after statistically controlling for the other variables in the model. The median logistic R was 1.448, and the median AR was .809, indicating that approximately 81% of the students were correctly classified by the

R1700


6

model as having declared a math-intensive

indicated that the effect of gender on STEM

STEM major.

major choice depended about students’ high school math and science GPAs. More

Results for Model 2 are presented in Table 4. As

specifically, the results suggested that the

in the first model, all of the student

gender effect decreased as STEM-HSGPA

characteristics were significantly related to

increased when holding all other predictors

STEM major choice. Having a tilt toward math

constant.3 The median logistic R and median

and science on the ACT-tilt and HSGPA-tilt

AR for Model 3 were essentially unchanged,

measures and having a tilt toward Things and

1.518 and .812, respectively, from those in Model

Ideas on the People/Things and Data/Ideas

2.

dimensions were each associated with an increased likelihood of declaring a STEM major. As these measures increased, the likelihood that a student would declare a math-intensive STEM major increased. Additionally, as was indicated in Model 1, results from Model 2 suggested that male students were more likely than female students to declare a STEM major (adjusted odds ratio = 1.21). However, the gender effect in Model 2 was somewhat smaller than it was in Model 1. This finding suggests that statistically controlling for the tilt measures helped to reduce but did not eliminate the gender gaps in STEM major choice. Adding the tilt measures slightly increased the model fit statistics: The median logistic R increased from 1.448 to 1.515 and the median AR increased from .809 to .812. Table 5 contains the results for the third model. Only one of the six interactions tested, Gender x STEM-HSGPA, was statistically significant. This finding suggests that the effect of the ACT STEM score and the four tilt measures on STEM major choice did not differ between males and females. For STEM-HSGPA, the interaction estimate was negative, indicating that the slope for STEM-HSGPA was steeper for female students than for male students. That is, the odds of declaring a STEM major that is associated with a one standardized unit increase in STEM-HSGPA was greater for females (adjusted odds ratio = 1.36) than for males (adjusted odds ratio = 1.24). Moreover, the significant Gender x STEM-HSGPA interaction

Conclusion In conclusion, the results from this study demonstrate the value of considering measures of academic tilt and vocational interest tilt when identifying whether a student is likely to enter a STEM major. Two measures for relative academic strength or tilt were explored that were derived by taking the differences in a student’s standardized STEM and non-STEM achievement levels. The first measure was based on ACT test scores and the second measure was based on subject area high school GPAs. The approach taken to obtain these tilt measures accounted for differences in the score distributions between STEM and non-STEM areas; this was done by first converting STEM and non-STEM achievement levels to standardized z-scores based on summary statistics obtained from a national reference group of ACT-tested students, and then subtracting the two zscores to obtain the academic tilt measures. Likewise, two measures of vocational interest tilt were examined using Prediger’s (1982) People/Things and Data/Ideas work-task dimension scores; these measures were standardized using summary statistics from the same national reference group. The two academic tilt measures were positively related to STEM major choice, meaning that students with greater tilt toward math and

R1700


7

science according to their test scores or their

variables in the model. At the outset of the

subject area high school GPAs had a greater

study, we had hypothesized that the gender

likelihood of declaring a math-intensive STEM

effect would disappear once the academic tilt

major. Additionally, as a student’s work-task

and vocational interest tilt measures were

dimension scores tilted more toward Things

taken into account. This did not occur, though

and Ideas on the People/Things and Data/Ideas

there was a reduction in the gender effect after

dimensions, the more likely they were to

the tilt measures were added to the model.

declare a STEM major. These results held even

Additionally, results from the third model that

after statistically controlling for mathematics

included interactions with gender indicated

and science academic achievement levels, high

that the magnitude of the gender effect on

school coursework taken and grades earned,

STEM major choice decreased as students’ high

major intentions, certainty of major intentions,

school mathematics and science GPA

and gender. The findings lend support to

increased. Future research is needed to better

previous research that has found a positive

understand this relationship, as well as to

relationship between having a tilt toward math

identify other factors, such as spatial ability

and science and choosing a STEM major (Coyle

(Andersen, 2014; Lubinski, 2010; Shea et al., 2001;

et al., 2014; Davison et al., 2014). This research

Snow, 1999; Super & Bachrach, 1957; Wai,

adds to our understanding of which students

Lubinski, & Benbow, 2009; Wood & Lebold,

are likely to declare a math-intensive STEM

1968) and academic climate (Flam, 1991; Gayles

major as compared to a non-STEM major.

& Ampaw, 2014; Walton, Logel, Peach, Spencer, & Zanna, 2014), that may explain the gender

Another key finding was that gender remained

gaps among students declaring a math-

an important predictor of STEM major choice

intensive STEM major in the first term of

after statistically controlling for the other

college.

R1700


8

References ACT. (2009). ACT interest inventory technical manual. Iowa City, IA: ACT. ACT. (2016). The condition of STEM 2016. Iowa City, IA: ACT. Allen, J., & Le, H. (2008). An additional measure of overall effect size for logistic regression models. Journal of Educational and Behavioral Statistics, 33(4), 416-441. Andersen, L. (2014). Visual–spatial ability: Important in STEM, ignored in gifted education. Roeper Review, 36(2), 114–121. Ceci, S. J., Williams, W. M., & Barnett, S. M. (2009). Women's underrepresentation in science: Sociocultural and biological considerations. Psychological Bulletin, 135(2), 218-261. Coyle, T. R., Purcell, J. M., Snyder, A. C., & Richmond, M. C. (2014). Ability tilt on the SAT and ACT predicts specific abilities and college majors. Intelligence, 46(1), 18-24. Davison, M. L., Jew, G. B., & Davenport, E. C. (2014). Patterns of SAT scores, choice of STEM major, and gender. Measurement and Evaluation in Counseling and Development, 47(2), 118-126. Flam, F. (1991). Still a ‘chilly climate’ for women? Science, 252(5013), 1604–1606. Gayles, J. G., & Ampaw, F. (2014). The impact of college experiences on degree completion in STEM fields at four-year institutions: Does gender matter? The Journal of Higher Education, 85(4), 439–468. Holland, J. L. (1959). A theory of vocational choice. Journal of Counseling Psychology, 6(1), 35– 45. Holland, J. L. (1997). Making vocational choices: A theory of vocational personalities and work environments (3rd ed.). Odessa, FL: Psychological Assessment Resources. Le, H., Robbins, S., & Westrick, P. (2014). Predicting student enrollment and persistence in college STEM fields using an expanded P-E fit framework: A large-scale multilevel study. Journal of Applied Psychology, 99(5), 915-947. Lubinski, D. (2010). Spatial ability and STEM: A sleeping giant for talent identification and development. Personality and Individual Differences, 49(4), 344–351. doi.org/10.1016/j.paid.2010.03.022 Lubinski, D. & Benbow, C. P. (2007). Sex differences in personal attributes for the development of scientific expertise In S. J. Ceci & W. M. Williams (Ed.), Why aren't more women in science? (pp. 79-100). Washington, DC: American Psychological Association. Mattern, K., & Radunzel, J. (2016, June). Who will declare a STEM major? The role of achievement and interests. Paper presented at Association for Institution Research Forum, New Orleans, LA. Mattern, K., Radunzel, J., & Westrick, P. (2015). Development of STEM readiness benchmarks to assist educational and career decision making. Iowa City, IA: ACT.

R1700


9

National Science Foundation, National Center for Science and Engineering Statistics. (2015). Women, minorities, and persons with disabilities in science and engineering: 2015 (Special Report NSF 15-311). Arlington, VA. Retrieved from http://www.nsf.gov/statistics/wmpd/. Pennock-Román, M. (1994). College major and gender differences in the prediction of college grades (College Board Research Report 94-2). New York, NY: The College Board. Prediger, D. J. (1982). Dimensions underlying Holland’s hexagon: Missing link between interests and occupations? Journal of Vocational Behavior, 21(3), 259 –287. Radunzel, J., Mattern, K., & Westrick, P. (2016). The role of academic preparation and interest on STEM success. Iowa City, IA: ACT. Riegle-Crumb, C., King, B., Grodsky, E., & Muller, C. (2012). The more things change, the more they stay the same? Prior achievement fails to explain gender inequality in entry into STEM college majors over time. American Educational Research Journal, 49(6), 1048-1073. Shea, D. L., Lubinski, D., & Benbow, C. P. (2001). Importance of assessing spatial ability in intellectually talented young adolescents: A 20-year longitudinal study. Journal of Educational Psychology, 93(3), 604–614. Snow, R. E. (1999). Commentary: Expanding the breadth and depth of admissions testing. In S. Messick (Ed.), Assessment in higher education: Issues of access, quality, student development, and public policy (pp. 133–140). Mahwah, NJ: Erlbaum. Su, R., Rounds, J., & Armstrong, P. I. (2009). Men and things, women and people: A meta-analysis of sex differences in interests. Psychological Bulletin, 135(6), 859-884. Super, D. E., & Bachrach, P. B. (1957). Scientific careers and vocational development theory. New York, NY: Bureau of Publications, Teachers College, Columbia University. Wai, J., Lubinski, D., & Benbow, C. P. (2009). Spatial ability for STEM domains: Aligning over 50 years of cumulative psychological knowledge solidifies its importance. Journal of Educational Psychology, 101(4), 817–835. dx.doi.org/10.1037/a0016127 Walton, G. M., Logel, C., Peach, J. M., Spencer, S. J., & Zanna, M. P. (2014). Two brief interventions to mitigate a “chilly climate” transform women’s experience, relationships, and achievement in engineering. Journal of Educational Psychology, 107(2), 468–485. dx.doi.org/10.1037/a0037461 Westrick, P. (2018). Examining precollege gender differences and similarities among fourth-year undergraduate students, with a focus on STEM. Iowa City, IA: ACT. Wood, D. A., & Lebold, W. K. (1968). Differential and overall prediction of academic success in engineering: The complementary role of DAT, SAT and HSR. Educational and Psychological Measurement, 28(4), 1223– 1228. doi.org/10.1177/001316446802800426

R1700


10

Notes 1.

See Table A1 from Radunzel et al. (2016) for a list of the individual Classification of Instruction codes included for each cluster.

2.

In the models, the major certainty variable was multiplied by a dichotomous variable denoting whether the student had an intended major (coded as 1) compared to undecided students (coded as 0) to ensure the inclusion of the undecided students in the STEM major choice models.

3.

For example, the gender coefficient was estimated to be 0.209 for students with a STEM HSGPA of 3.15 (adjusted odds ratio = 1.23), 0.171 for a student with a STEM HSGPA of 3.41 (adjusted odds ratio = 1.19), and 0.091 for a student with a STEM HSGPA of 3.98 (adjusted odds ratio = 1.09), holding all other predictors constant at their sample mean values.

Figure 1. Third edition of the ACT World-of-Work Map (ACT, 2009)

R1700


11

Tables Table 1. Descriptive Statistics for the 2003-2009 National ACT-Tested Population Overall Observed

N

Male

Mean

SD

N

Female

Mean

SD

N

Mean

SD

4.4

ACT STEM Score

7,767,175

20.9

4.7

3,380,370

21.4

4.9

4,207,731

20.4

ACT English and Reading Score

7,767,175

21.0

5.7

3,380,370

20.6

5.8

4,207,731

21.2

STEM HSGPA

5,975,818

3.15

0.68

2,536,262

3.09

0.71

3,361,215

3.19

0.66

English and Social Studies HSGPA

6,030,699

3.30

0.64

2,559,982

3.20

0.67

3,391,648

3.38

0.60

People-Things

6,485,296

-0.07

32.35

2,781,554

9.44

31.32

3,610,212

-7.36

31.23

Data-Ideas

6,485,296

-3.48

33.75

2,781,554

-2.84

32.56

3,610,212

-3.97

34.64

High School Calculus*

5,954,481

0.13

0.34

2,527,008

0.14

0.35

3,349,426

0.12

0.32

High School Physics*

6,192,666

0.33

0.47

2,647,893

0.37

0.48

3,458,229

0.31

0.46

ACT STEM Score

7,767,175

0.00

1.00

3,380,370

0.12

1.06

4,207,731

-0.10

0.94


7,767,175

0.00

1.00

3,380,370

-0.06

1.01

4,207,731

0.04

0.99

STEM HSGPA

5,975,818

0.00

1.00

2,536,262

-0.09

1.03

3,361,215

0.06

0.97


6,030,699

0.00

1.00

2,559,982

-0.17

1.06

3,391,648

0.12

0.94

People/Things

6,485,296

0.00

1.00

2,781,554

0.29

0.97

3,610,212

-0.23

0.97

Data/Ideas

6,485,296

0.00

1.00

2,781,554

0.02

0.96

3,610,212

-0.01

1.03

ACT-Tilt**

7,767,175

0.00

0.63

3,380,370

0.18

0.62

4,207,731

-0.14

0.60

HSGPA-Tilt**

5,897,831

0.00

0.69

2,500,348

0.07

0.70

3,320,751

-0.07

0.67

5.7

Standardized

Notes. *The mean is the proportion of students who reported taking the course. **Calculated using the standardized measures, hence the SDs do not necessarily equal 1. SD = Standard deviation.

R1700


12

Table 2. Descriptive Statistics for the Overall Sample Overall Observed

N

Mean

ACT STEM Score

79,464

22.7


79,464

STEM HSGPA

79,464


Male SD

N

Mean

4.6

35,684

23.5

23.2

5.5

35,684

22.9

3.41

0.57

35,684

3.37

78,854

3.56

0.49

35,373

3.48

People/Things

75,510

-2.84

33.79

33,622

Data/Ideas

75,510

-2.69

35.77

33,622

High School Calculus*

79,464

0.20

0.40

35,684

High School Physics*

79,464

0.39

0.49

ACT STEM Score

79,464

0.40


79,464

STEM HSGPA

79,464


Female SD

N

Mean

SD

4.8

43,780

22.1

4.2

5.6

43,780

23.4

0.59

43,780

3.45

0.54

0.52

43,481

3.62

0.45

6.87

32.56

41,888

-10.63

32.72

-3.01

34.18

41,888

-2.43

36.99

0.22

0.42

43,780

0.18

0.38

35,684

0.42

0.49

43,780

0.36

0.48

0.98

35,684

0.56

1.04

43,780

0.27

0.91

0.38

0.95

35,684

0.34

0.97

43,780

0.42

0.93

0.39

0.83

35,684

0.33

0.87

43,780

0.44

0.79

78,854

0.40

0.77

35,373

0.28

0.82

43,481

0.49

0.70

People/Things

75,510

-0.09

1.04

33,622

0.21

1.01

41,888

-0.33

1.01

Data/Ideas

75,510

0.02

1.06

33,622

0.01

1.01

41,888

0.03

1.10

ACT-Tilt

79,464

0.02

0.66

35,684

0.22

0.64

41,888

-0.33

1.01

HSGPA-Tilt

78,854

-0.01

0.57

35,373

0.05

0.58

41,888

0.03

1.10

5.3

Standardized

Note. *The mean is the proportion of students who reported taking the course. SD = Standard deviation.

R1700


13

Table 3. Predicting Math-Intensive STEM Major Choice, Model 1 Solutions for Fixed Effects Intercept

Estimate -2.412

SE

DF

t Value

0.110

42

-21.87

p Value